It doesn't seem like it; check out the adversarially constructed images here. They don't look anything like the original despite perfectly matching the NeuralHash: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
Think more like a “deep dream” than these adversarial attacks.